6
$\begingroup$

I have done a ncvTest, but I am not really sure how to interpret this properly. I look documentation and examples online but was not able to find anything that clearly explains about ncvTest.

Here are my results from ncvTest:

Non-constant Variance Score Test 
Variance formula: ~ fitted.values 
Chisquare = 3.203987    Df = 1     p = 0.073459

enter image description here

Improve this question
$\endgroup$

1 Answer 1

Reset to default
6
$\begingroup$

This test is more prominently know as Breusch-Pagan test. It is a test for heteroscedasticity. In a standard linear model, the variance of the residuals are assumed to be constant (i.e. independent) over the values of the response (fitted values).

In your specific case, there is some evidence for a non-constant variance of the residuals (heteroscedasticity). A good suggestion would be to plot the residuals vs. the fitted values which you can do in R with plot(reg.mod) after calculating the regression model.

Also, have a look at the search page with the term "Breusch-Pagan" of cross validated. There are a lot of questions similar to yours.

Improve this answer
$\endgroup$
3
  • $\begingroup$ Yeah I added the model plots my regression is Y ~ log(X) but I don't think this is a good model? I am trying to see how ncvTest can tell me that. $\endgroup$
    – add-semi-colons
    May 14, 2013 at 9:22
  • 4
    $\begingroup$ Diagnostics are good, but the one piece of evidence that is even better is a scatter plot showing your two variables and the fitted line! I count here 24 or so values; with that small a sample size it's easy to over-interpret diagnostics. What seems most obvious is that you have two clusters and you're fitting a line through them. Whether that is a good idea is more a matter of the science underlying your data than something on which someone statistical can proclaim either blessing or curse. Whether it is a good model depends on whether you can think of a better one. $\endgroup$
    – Nick Cox
    May 14, 2013 at 9:33
  • 1
    $\begingroup$ The residuals seem to be hinting at some curvature not quite captured. I'd throw at this more exploration, not more testing. $\endgroup$
    – Nick Cox
    May 14, 2013 at 9:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.